Address Copilot review: add flat->tensor fan-out to DDt fallback

Copilot flagged that after the shape fix, the multicomponent projector
writes into _psi_star_flat_var but update_pre_solve() never unpacks the
solved flat (1, Nc) field back into psi_star[0]. Mirrors the fan-out
already in initialise_history() (~ddt.py:1540): walk
_psi_star_indep_indices, copy flat var k-th component into
psi_star[0][:, i, j], and symmetric-fill [:, j, i] for SYM_TENSOR.

In practice the SemiLagrangian advection step at line 1016 then
modifies psi_star[0] in place, so the missing fan-out wasn't producing
an immediate symptom in this short test — but the projected flux value
was being silently discarded. Fix restores the intended semantics:
psi_star[0] reflects the current psi_fn (projected flux) before SL
advection updates it for the upstream sample.

Reverted the over-eager second test (asserting psi_star[0] non-zero
after one solve) since the SL advection masks the missing fan-out in
that observation window. Kept the ShapeError regression test which
remains a definitive guard for the original visible failure.

Underworld development team with AI support from Claude Code

Author: lmoresi

src/underworld3/systems/ddt.py

@@ -1776,6 +1776,17 @@ def update_pre_solve(
                 self._psi_star_projection_solver.smoothing = 0.0
                 self._psi_star_projection_solver.solve(verbose=verbose)
 
+                # For tensor vtypes the projection writes into the flat (1, Nc) variable,
+                # so we must fan it back out to psi_star[0] — otherwise subsequent
+                # history operations read a stale tensor. Mirrors the same fan-out in
+                # initialise_history() (~line 1540).
+                if getattr(self, '_psi_star_use_multicomponent', False):
+                    for k, (i, j) in enumerate(self._psi_star_indep_indices):
+                        vals = self._psi_star_flat_var.array[:, 0, k]
+                        self.psi_star[0].array[:, i, j] = vals
+                        if i != j:
+                            self.psi_star[0].array[:, j, i] = vals
+
         # 3. Compute the upstream values from the psi_fn
 
         # We use the u_star variable as a working value here so we have to work backwards

tests/test_0610_navier_stokes_slcn_projection.py

@@ -1,22 +1,18 @@
 """
-Regression test for the NavierStokesSLCN DFDt projection source shape.
+Regression test for the NavierStokesSLCN DFDt projection fallback.
 
 When ``SNES_MultiComponent_Projection`` was wired into the ``SemiLagrangian``
 DDt path, the ``psi_fn`` setter and ``_setup_projections`` were updated to
 flatten the source tensor to a ``(1, Nc)`` row matrix via
 ``_build_projection_source``. The fallback path inside ``update_pre_solve``
 (taken when ``uw.function.evaluate`` raises on expressions containing
-derivatives — which is the NavierStokes viscous flux every step) missed
-the migration and assigned ``self.psi_fn`` directly, producing:
+derivatives — true for the NavierStokes viscous flux every step) missed the
+migration and assigned ``self.psi_fn`` directly, producing:
 
     sympy.matrices.exceptions.ShapeError:
         Matrix size mismatch: (1, 3) + (2, 2).
 
 See: https://github.com/underworldcode/underworld3/issues/180
-
-The test does one ``solve(timestep=dt)`` of NavierStokesSLCN on a tiny mesh.
-Before the fix this raises ``ShapeError`` from the DDt fallback. After the
-fix it converges normally.
 """
 import pytest
 import sympy
@@ -26,6 +22,7 @@
 @pytest.mark.level_2
 @pytest.mark.tier_a
 def test_navier_stokes_slcn_solve_does_not_raise_shape_error():
+    """First solve completes without ShapeError from the DDt projection fallback."""
     mesh = uw.meshing.UnstructuredSimplexBox(
         minCoords=(0.0, 0.0),
         maxCoords=(1.0, 1.0),